Bringing AI to the Most Important 5 Minutes in League of Legends

How we built an intelligent draft assistant — and what it taught us about decision-making under pressure

Introduction: The Game Before the Game

In professional League of Legends, the outcome of a match is often decided before a single minion spawns.

The draft phase — those five intense minutes of champion select — is where games are won or lost. Both teams take turns banning champions they don’t want to face and picking champions they want to play. It’s a high-stakes chess match with incomplete information, time pressure, and consequences that ripple through the next 30–45 minutes of gameplay.

If you draft well, your players have tools to succeed. The right champions to counter opponents. Synergies that amplify each other. Win conditions that are achievable.

If you draft poorly? Your players are fighting uphill from minute one. Bad matchups. Awkward compositions. Win conditions that are theoretical at best.

Professional teams spend enormous resources on draft preparation. Coaches spend hours analyzing opponent tendencies, champion statistics, and meta trends. Players develop deep champion pools to give their teams flexibility. Entire practice sessions are devoted to specific compositions.

And yet, when the clock is ticking in champion select, decisions often come down to instinct, memory, and best guesses.

What if AI could help?

The Chaos of Champion Select

If you’ve never experienced professional League of Legends champion select, let me paint the picture.

The room is tense. Players sit at their computers, ready to execute whatever the coaching staff decides. The head coach stands behind them, communicating through headsets. A timer counts down — you have limited time to lock in each selection.

On screen, the draft interface shows:

• The champions available to pick
• The champions already banned
• The champions already picked by both teams
• A ticking clock

Now the decisions begin.

“They banned Maokai. What does that tell us about their plans?” “Should we pick Orianna now or wait to see their mid-pick first?” “If we take this champion, what does that leave them for response picks?”

There are over 160 champions in League of Legends. Each one has unique abilities, strengths, and weaknesses. Some champions counter others. Some synergize beautifully. Some are powerful in the current meta; others are weak.

The number of possible 10-champion draft compositions is astronomical — literally millions of unique combinations. No human being can hold all of this information in their head while also managing time pressure, reading opponents, and coordinating with their team.

Currently, teams prepare “scripts” — pre-planned draft paths that they’ve practiced and theorycrafted. But scripts can only take you so far. Every draft is different. Opponents surprise you. The meta shifts. You need to adapt in real-time.

That’s where Cloud9 Draft Assistant comes in.

Introducing Cloud9 Draft Assistant

Cloud9 Draft Assistant is an AI-powered draft companion for League of Legends that provides real-time recommendations during champion select.

It doesn’t replace coaches — it augments them. While the humans focus on reading opponents, managing team communication, and making final decisions, the AI handles:

• Tracking the current draft state
• Recommending optimal picks and bans
• Predicting what opponents might select
• Calculating win probability as the draft unfolds
• Explaining the reasoning behind each recommendation

Think of it like a chess computer that suggests moves, explains why, and shows how each decision affects your chances of winning — except for League of Legends champion select.

Feature Deep Dive: What Can It Do?

The Draft Board

The centerpiece is the Draft Board — a visual interface that mirrors the actual champion select screen.

On the left, blue team’s bans and picks. On the right, red team’s bans and picks. In the center, a searchable grid of all available champions. At the top, the current phase indicator: BAN 1, PICK 1, BAN 2, etc.

As each selection happens in the real game, you (or an assistant) log it in the Draft Board. The interface updates instantly, showing what’s been selected and what’s still available.

But the Draft Board isn’t just for tracking — it’s actively analyzing.

AI Recommendations

For each phase of the draft, the AI provides recommendations.

Let’s say it’s Blue Team’s first pick. The AI might recommend:

Recommended: Maokai Confidence: High Reasoning: “Maokai is the top priority support in the current meta with an 54% professional win rate. Based on opponent tendencies, they’re likely to pick Nautilus if Maokai is available — Maokai is stronger in head-to-head matchups. Additionally, Maokai’s crowd control synergizes well with later pick options like Orianna or Jinx.”

Alternative: K’Sante Confidence: Medium Reasoning: “K’Sante is a flex pick that can go top or mid, hiding your strategy. Taking K’Sante early keeps more options open for later phases.”

Alternative: Nautilus Confidence: Medium Reasoning: “If you’re confident Maokai will be banned by the opponent, Nautilus is a comparable option with strong engage.”

The AI doesn’t just say “pick this champion.” It explains why — considering synergy, counterplay, meta strength, and opponent tendencies. Coaches can take the recommendation at face value or use the reasoning to inform their own decision.

Win Probability Meter

Throughout the draft, a win probability meter displays the current estimated chance of victory for each team.

The draft starts at 50/50. As selections happen, the probability shifts:

Blue picks Maokai: 52% Blue — 48% Red Red picks Nautilus: 51% Blue — 49% Red Blue picks Orianna: 53% Blue — 47% Red Red picks Yone: 54% Blue — 46% Red

Players and coaches can watch the probability update in real-time, understanding how each decision affects their chances. Did that pick help or hurt? How much? The meter provides instant feedback.

This is powerful for learning. After the match, teams can review the draft and see exactly when their probability spiked or dipped. “That’s when we made a mistake” becomes obvious.

Opponent Predictions

One of the hardest parts of the draft is anticipating what the opponent will do.

Based on historical data from GRID’s professional match database, the AI predicts likely opponent selections:

“Based on their draft tendencies, the opponent’s likely next pick is:

• Kai’Sa (45% probability) — their ADC has 80% pick rate on this champion
• Varus (30% probability) — fits their current composition
• Ezreal (15% probability) — safe fallback option”

This information is invaluable. If you know the opponent is likely to pick Kai’Sa, you can prepare a counter — either banning her or picking a champion that matches well against her.

Prediction isn’t perfect — opponents sometimes surprise you — but having a data-driven estimate is far better than pure guesswork.

Draft Simulation

What if you want to practice drafts without the pressure of a real match?

Draft Simulation mode lets you run practice drafts against an AI opponent. You pick for one team; the AI responds intelligently for the other.

This is useful for:

• Testing new strategies before using them in competition
• Exploring how different draft paths feel
• Training new players or coaches on draft principles
• Preparing for specific opponent tendencies

The AI opponent isn’t perfect — it won’t perfectly replicate any specific team — but it provides reasonable resistance that makes practice meaningful.

Meta Analysis

The meta in League of Legends shifts constantly. Patches change champion power levels. Professional play discovers new strategies. What was weak last month might be dominant today.

The Meta Champions feature provides a current tier list based on recent professional play data:

S-Tier (Always Pick/Ban):

• Maokai (62% pick/ban rate, 54% win rate)
• K’Sante (85% pick/ban rate, 52% win rate)
• Orianna (58% pick/ban rate, 51% win rate)

A-Tier (Strong Options):

• Nautilus, Rell, Jinx, Xayah, Azir…

B-Tier (Situational):

• …and so on

This keeps teams informed about the current meta without requiring hours of research. When the patch changes and the meta shifts, the data updates accordingly.

Real Scenario: How It Works in Practice

Let’s walk through a realistic draft scenario to see how Cloud9 Draft Assistant adds value.

The Setup: Your team (Blue side) is facing a strong opponent in a playoff match. You’ve prepared several strategies but need to adapt based on what the opponent shows.

Ban Phase 1:

You ban Maokai — a power pick you don’t want to face. Opponent bans K’Sante. You ban Yone — their mid-laner’s signature champion. Opponent bans Jinx.

The AI notes: “Opponent banning Jinx but not their own strong ADC targets suggests they plan to play around ADC. Consider prioritizing bot pressure.”

Pick Phase 1:

It’s your first pick. The AI recommends Nautilus:

“With Maokai banned, Nautilus is the next-best support option. Taking it denies the opponent their likely support plan while giving you strong engage. Win probability increases to 52% if selected.”

You pick Nautilus.

Opponent picks Viego and Xayah. A jungle-ADC core that signals their composition direction.

The AI updates: “Opponent showing AD-heavy composition. Consider AP mid and tank top to balance damage profiles. Their Viego is vulnerable to crowd control chains.”

Pick Phase 2:

Your turn for two picks. The AI recommends:

1. Orianna (mid) — “Strong control mage with CC that chains well with Nautilus. Good matchup into likely opponent mid picks.”
2. Sejuani (jungle) — “Another CC threat that synergizes with your composition. Makes Viego’s life difficult.”

You lock in Orianna and Sejuani.

Win probability: 55% Blue.

And so on…

Throughout the draft, the AI provides continuous guidance. Not making decisions for you — but ensuring you have the best possible information to make your own decisions.

The Philosophy: Augmentation, Not Replacement

A common fear with AI in any competitive domain is that it will replace human judgment. Coaches might worry: “If the AI is making recommendations, why do you need me?”

This concern misses how the tool is designed.

Cloud9 Draft Assistant augments coaches. It handles the computational complexity — tracking champion relationships, calculating probabilities, analyzing opponent data — so coaches can focus on what humans do best:

• Reading the room. Is our star player confident on this champion? Is the opponent’s coach showing nervousness?
• Managing egos. Sometimes the “optimal” pick isn’t the right pick because of team dynamics.
• Long-term thinking. Do we want to reveal this strategy now, or save it for a more important match?
• Intuition. Sometimes experienced coaches have a gut feeling that defies the data — and they’re often right.

The AI provides information. Humans provide wisdom. Together, they make better decisions than either could alone.

The Data Behind the Recommendations

You might wonder: where do these recommendations come from? How does the AI “know” what to suggest?

The answer is data — specifically, professional match data from GRID.

We analyze thousands of professional League of Legends games to understand:

Champion Performance: Raw statistics on win rates, pick rates, ban rates, and performance metrics across different skill levels and regions.

Synergies: Which champion combinations perform better than expected when played together. Orianna + Gnar, for example, have a famous synergy around combined ultimate abilities.

Counters: Which champions perform well against which other champions. Some matchups are so lopsided that picking into them is a significant mistake.

Opponent Tendencies: How specific teams and players behave in draft. Some players have very narrow champion pools; others are flexible. Some teams always prioritize certain strategies.

All of this data flows into the recommendation engine, which synthesizes it in real-time during the draft.

Lessons from Building the Draft Assistant

Creating Cloud9 Draft Assistant taught us about competitive gaming, AI, and the nature of good decision support.

Lesson 1: League of Legends Draft Is Incredibly Complex

When we started, we underestimated how complex champion select really is. 160+ champions. Role assignments. Synergies. Counters. Meta shifts. Opponent habits. Team preferences. Player champion pools. Flex picks that hide information.

Building a system that could reason about all of this required serious effort. We have new respect for professional coaches who manage this complexity without AI assistance.

Lesson 2: Explanation Matters as Much as Recommendation

Early versions of the tool just said: “Pick Orianna.” Users hated it. They didn’t trust recommendations they couldn’t understand.

When we added explanations — “Pick Orianna because she synergizes with Nautilus and counters likely enemy mid picks” — trust increased dramatically. Users felt like they were getting useful information, not mysterious commands.

This is a general lesson for AI tools. Transparency builds trust.

Lesson 3: Speed Is Non-Negotiable

In champion select, you have limited time for each decision. An AI that takes 30 seconds to generate a recommendation is useless — the clock has already moved on.

We optimized aggressively for speed. Recommendations appear in under a second. The trade-off was sometimes accepting slightly less comprehensive analysis, but the speed was worth it.

What’s Next for Draft?

We’re excited about future development:

Live Integration with Professional Broadcasts: Imagine watching a professional match and seeing AI recommendations in real-time, just like the teams might see. Great for fans who want to understand draft strategy.

Team-Specific Training: Custom AI models trained on your specific team’s data, understanding your player pools and strategic preferences.

Voice Interface: Hands-free operation so coaches can receive recommendations without looking at screens.

Post-Draft Analysis: Detailed breakdowns after the draft showing key decision points and what alternatives existed.

Why This Matters: The Democratization of Intelligence

The best esports organizations have analytics departments. They have coaches with decades of experience. They have resources to spend on preparation.

Smaller teams don’t have those advantages. They’re competing with fewer resources, less experience, and less information.

AI tools like Cloud9 Draft Assistant level the playing field. The analytical power that used to require a dedicated analyst is now available to everyone. The pattern recognition that used to require years of experience is now embedded in software.

This democratization is good for esports. More competitive matches. More upsets. More storylines. The dominance of resource-rich organizations becomes less assured when everyone has access to intelligent tools.

Conclusion: Drafting Smarter

The draft phase will always be a human endeavor. The decisions, the pressure, the moments of insight — these are what make champion select exciting.

But humans don’t have to do it alone.

Cloud9 Draft Assistant provides the information backbone that enables better decisions. Track the draft state. Get recommendations with explanations. Watch win probability shift. Predict opponent moves.

The game within the game just got more intelligent.

Built for the Cloud9 x GRID Hackathon. We’d love to hear your thoughts!

Try It Yourself:

• Live Demo: c9-draft-ui.vercel.app
• API Documentation: c9-draft-api.onrender.com/docs
• GitHub: github.com/AvishKaushik/c9-draft-ui